sensitivity analysis

Use this forum to post Vensim related questions.
Post Reply
Nesli
Member
Posts: 21
Joined: Tue Jan 18, 2011 11:33 am

sensitivity analysis

Post by Nesli »

Hi,

I am using Vensim PLE plus and want to conduct sensitivity analysis for one parameter (C) by setting a range for two parameters (A and B)..

However parameters A and B are not in a feedback loop with parameter C..Their relation is like below:

C= X+Y*B
Y=W*A

Can I still use sensitivity analysis to figure out the effect of this change on parameter C?

Regards
tomfid
Administrator
Posts: 3804
Joined: Wed May 24, 2006 4:54 am

Re: sensitivity analysis

Post by tomfid »

Yes - Vensim doesn't care whether the inputs and outputs of a sensitivity analysis are related in any particular way.
Nesli
Member
Posts: 21
Joined: Tue Jan 18, 2011 11:33 am

Re: sensitivity analysis

Post by Nesli »

Thank you,

Then it means, for example SALES model (in Vensim Guide) if we set a range for "price of item" (lets say 100-200 random unifrom), we can run sensitivity for cash flow..

When I do so, the histogram for the last time appears with two peak columns (bimodel)..
However I know that more often, we tend to get a normal type distribution where a single peak displays the most frequently occurring hence which has the largest probability of occurrence. This clustering around a single value of cash flow and then a distribution of other values with smaller frequencies tailing off the further away from the peak frequency is known as a unimodel distribution.

Is there any particular reason of having bimodel type of behaviour??..I am asking this because I am having biomodal histogram in my model when I do sensitivity analysis and trying to understand why (btw I have no sinusoidal input) (and I tried to increase the number of simulations)..
Thanks
Administrator
Super Administrator
Posts: 4573
Joined: Wed Mar 05, 2003 3:10 am

Re: sensitivity analysis

Post by Administrator »

Can you post anything for us to take a look at? It's difficult to comment unless we can see the model/sensitivity settings.

Tony.
Advice to posters seeking help (it really helps us to help you)
http://www.ventanasystems.co.uk/forum/v ... f=2&t=4391

Units are important!
http://www.bbc.co.uk/news/magazine-27509559
tomfid
Administrator
Posts: 3804
Joined: Wed May 24, 2006 4:54 am

Re: sensitivity analysis

Post by tomfid »

I assume this is the sales.mdl in Chapter 15 of the user guide.

This could just be an artifact of the aggregation in the bar chart. Do you see the same behavior if you look at individual traces, or change the settings on the bar graph tool?
Nesli
Member
Posts: 21
Joined: Tue Jan 18, 2011 11:33 am

Re: sensitivity analysis

Post by Nesli »

Thank you,

I simplified my model and attached.
What I want to do is to run a sensitivity test for B by setting these ranges to my two parameters:
Y=RANDOM UNIFORM (1,1.8)
D= RANDOM UNIFORM (0.05-0.20)

Regardless of the number of simulation I have two peaks in histogram (for the final time)..

I am wondering whether it is an artifact of the aggregation in the bar chart ? How can I understand the reason of having this kind of bar chart?

regards
Attachments
nesli1.mdl
(3.05 KiB) Downloaded 214 times
Nesli
Member
Posts: 21
Joined: Tue Jan 18, 2011 11:33 am

Re: sensitivity analysis

Post by Nesli »

Y= RANDOM UNIFORM (1, 1.8 )
Administrator
Super Administrator
Posts: 4573
Joined: Wed Mar 05, 2003 3:10 am

Re: sensitivity analysis

Post by Administrator »

What parameters do you see the two peaks?

The plots from this all look fine to me.
pic1.jpg
pic1.jpg (46.43 KiB) Viewed 8142 times
pic2.jpg
pic2.jpg (59.48 KiB) Viewed 8142 times
Advice to posters seeking help (it really helps us to help you)
http://www.ventanasystems.co.uk/forum/v ... f=2&t=4391

Units are important!
http://www.bbc.co.uk/news/magazine-27509559
Nesli
Member
Posts: 21
Joined: Tue Jan 18, 2011 11:33 am

Re: sensitivity analysis

Post by Nesli »

Could you pls check bar chart histogram for the final time..there are two peak columns..
Nesli
Member
Posts: 21
Joined: Tue Jan 18, 2011 11:33 am

Re: sensitivity analysis

Post by Nesli »

I managed to attach the bar chart as Jpeg..
It is a histogram for parameter B..
More often, you tend to get a normal type distribution where a single peak displays the most frequently occurring. This clustering around a single value of B and then a distribution of other values with smaller frequencies..But in this case it is different than this..Is that normal to have two peaks like this?

Thanks
Attachments
nesli3.JPG
nesli3.JPG (43.24 KiB) Viewed 8140 times
Administrator
Super Administrator
Posts: 4573
Joined: Wed Mar 05, 2003 3:10 am

Re: sensitivity analysis

Post by Administrator »

Are you sure the histogram should be displaying a normal distribution? In your model, "e" and "c" could easily cause the two peaks.
Advice to posters seeking help (it really helps us to help you)
http://www.ventanasystems.co.uk/forum/v ... f=2&t=4391

Units are important!
http://www.bbc.co.uk/news/magazine-27509559
Nesli
Member
Posts: 21
Joined: Tue Jan 18, 2011 11:33 am

Re: sensitivity analysis

Post by Nesli »

I thought that C should be decreasing as I set a range for Y smaller than its actual value so i thought only E*D could cause an increase for B as I set a range for D bigger than its actual value..

Could you please explain me more about what I think please-I understand your comment for E but how does C cause peak??

Many thanks in advance
tomfid
Administrator
Posts: 3804
Joined: Wed May 24, 2006 4:54 am

Re: sensitivity analysis

Post by tomfid »

I think what you're seeing may simply be noise rather than real peaks. The input Y and D distributions look fairly uniform, so any peaks that emerge should be from the interactions in your system rather than the random number generation.

The sensitivity histograms in Vensim may not make ideal binning choices, so if you really want to know what's going on, you should export your simulation output to another tool for visualization.
Nesli
Member
Posts: 21
Joined: Tue Jan 18, 2011 11:33 am

Re: sensitivity analysis

Post by Nesli »

Thank you so much for your response..

I exported the sensitivity run to the excel file and traced the values..

When I set D in a range of 0.05-0.45 rather than (0.05-0.20) I have one peak or if I use Random Normal Simulation I have one peak again..

If it is not too many questions can I ask;

(1) The figures in Random Normal Distribution is in a form of Bell Curve with mean and standart deviation(thats fine), but in random uniform how the values in a range corresponds to the each simulation? (is it something like multiplicative congruential algorithm?)

(2) You say the peaks might not be peaks but noice..Does that mean they are randomly highest frequencies rather than peaks?

sorry if I bother with lots of emails..
thanks a lot..
tomfid
Administrator
Posts: 3804
Joined: Wed May 24, 2006 4:54 am

Re: sensitivity analysis

Post by tomfid »

1. RANDOM UNIFORM is a uniform distribution over the specified range (i.e. probability of a draw inside the range is the same everywhere, and 0 outside the range).

2. Correct. You could test this by repeating your experiments with different random number seeds and/or a larger sample size.
Nesli
Member
Posts: 21
Joined: Tue Jan 18, 2011 11:33 am

Re: sensitivity analysis

Post by Nesli »

OK thank you..
Post Reply